TinyML-Driven Cybersecurity for Autonomous Spacecraft: Latency-Accuracy Analysis for SPARTA RF and Cyber Threat Detection

Researchers demonstrate that lightweight machine learning models, particularly Logistic Regression, can detect cyber and RF threats on autonomous spacecraft with microsecond-level inference speeds and minimal accuracy loss compared to more complex models. The study analyzes TinyML-compatible algorithms against the SPARTA attack model, showing practical feasibility for real-time onboard threat detection in resource-constrained space environments.

This research addresses a critical vulnerability in autonomous spacecraft systems: the need for rapid, lightweight cyber threat detection without relying on ground-based monitoring or large computational resources. As space operations increasingly depend on autonomous decision-making, the ability to detect and respond to RF jamming, spoofing, and command injection attacks in real-time becomes essential for mission continuity and national security.

The advancement builds on broader trends in edge computing and embedded AI, where resource constraints demand algorithmic efficiency without sacrificing detection capability. The SPARTA attack model provides a comprehensive threat taxonomy for spacecraft systems, and the researchers' physics-informed analysis of model complexity demonstrates how classical ML approaches can outperform deep learning for specific space applications. This contrasts with typical industry trends favoring neural networks, highlighting that domain-specific constraints sometimes favor simpler, more interpretable solutions.

For the space and defense sectors, this research validates TinyML approaches for autonomous systems operating in contested environments where communication delays and hardware limitations are fundamental constraints. The finding that Logistic Regression achieves near-parity with Random Forest while reducing latency dramatically improves deployment feasibility for satellite operators and spacecraft developers. This enables faster iteration cycles and reduces the computational burden on increasingly autonomous systems.

Future developments will likely focus on hybrid architectures combining multiple timescale detectors and enhanced feature extraction methods. The research opens opportunities for specialized hardware implementations and federated learning approaches across satellite constellations. As space becomes more commercialized and contested, these cybersecurity capabilities represent strategic differentiators for operators managing critical infrastructure dependencies.

• →Logistic Regression achieves microsecond-level inference with only 1% accuracy loss compared to Random Forest for spacecraft threat detection
• →TinyML models effectively detect SPARTA-class attacks including jamming, spoofing, payload manipulation, and unauthorized commands
• →Physics-informed theoretical analysis of computational complexity and latency scaling validates model selection for resource-constrained spacecraft
• →Classical ML approaches outperform deep learning for this domain due to hardware constraints and real-time requirements
• →The research enables practical onboard autonomous threat detection without relying on ground-segment communication